Struggling with control points

 I’ve read the documentation and some help topics but still not clear…

I’m making a detailed model from a mixture of cameras. In normal detail it comes out to 240m or so triangles.

Some of the textures are exactly what I want…others are “fuzzy”, “blobby” or smeared…or in some cases just plain blank. I have enough photos to cover the entire model, so I’m guessing this is an alignment issue…

From what I understand, I can add a control point manually in multiple images that represents the same physical point in space and RC will “rethink” it’s alignment.

When it works (and it has), that’s great - it reduces the number of components and visually produces much crisper results.

It’s when it doesn’t work that I’m not clear how to address…when I get those little yellow or red flags and a “px” difference that is higher than 1 or 2 (depending on my minimum error settings).

I’ve turned residuals on in the “Image” view, so I can see the orange line that points to where RC thinks the control point should be. If the new destination looks sensible (sometimes I get it wrong, or the resolution on the photographs is insufficent to be ENTIRELY sure) then that’s easy…just nudge the control point over to the other end of the orange line.

But what about when you have 39 instances of the same control point all with “px” values below 1.6, except for one that is 3.9 - and when you look at it it’s just obviously, clearly wrong to move it to the suggested destination?

Do I move the point and trust RC?
Do I leave the point where it is (in which case I think RC will ignore it)?
Do I delete the point entirely as it’s just causing confusion?

In general with control points I have the following questions;

How many control points is a minimum/maximum or “best” number?
In how many images should each control point be referenced? (again, min, max, best)
With the px values, is it worthwhile (or even possible) to spend time reducing the px values to 0.0 for all of them?

Thanks!

Pic to show what I mean…

Hello,

the yellow triangles have nothing to do with control points: https://support.capturingreality.com/hc/en-us/articles/115001530852-Yellow-triangle-with-an-exclamation-mark

The red triangles should be treated – either corrected or removed. Sometimes when you remove one, more of them disappear, so this is quite fast.

Recommendations to the number of control points can be found in the Help section of the software. However, it depends on the dataset. Too many control points can lead to crashes.

I am afraid you cannot get to eliminating the error to 0, so there is no point in trying to achieve that. Just keep it in the default range.

Lucia - sorry but it’s not true that yellow triangles don’t relate to control points. The one your article talks about do NOT, because they are against the “Images” heading…but the ones in the control points list are NOT about missing camera information…they are a “could do better” warning, rather than an “unacceptable” warning.

If you don’t believe me you can prove it to yourself…if you move the cvonttrol point with a yellow warning to reduce the px error (ie in the direction of the orange line) - the yellow triangle goes away.

Roughly, px>1.5 = yellow triangle, px>2.0 = red triangle - though it varies.

The px is I guess the distance from the theoretical position of the control point, which is why if you move one, it can affect all the others (cos the “target” moves too…)…

So…thanks…and yes thanks for the info on numbers of points…I think though I may just be pushing the limits of what RC can do…(plus, very new at this so who knows what OTHER mistakes I am making in the background!)

you are right, I forgot about the ambigious yellow icon

Hi Nigel,

yes, it is absolutely possible to push the errors to 0.0 px error, although not really neccessary. How low depends on your needs. I often get errors of 0.3 or lower without much effort.

The point is (pun intended), that the error points to a difference in what you see and what RC calculates. So moving the CP to where RC says it should be can actually be counterproductive since it might be wrong and you correct. What I do in such cases is that I delete all components and start from scratch. Because if you keep some components RC will go from there and only nudge a bit but it won’t entirely rethink everything.

 “I delete all components and start from scratch. Because if you keep some components RC will go from there and only nudge a bit but it won’t entirely rethink everything.”

That’s wonderfully clear - is it really as good as that? I haven’t seen anyone say it like that.

Hi Tom,

I would say so.

I feel like I’ve been saying it in many threads over the past year…   :wink:

Force Component Rematch = True means the same thing, only that deleting older components is a more direct way of achieving the same goal.

That’s another super-clarifying statement!

If only all such insights could be made explicit, RC would look a lot simpler!

Putting Götz Echtenacher’s idea into practice as we speak. I’ll admit I kept the “best” component as I thought that would help RC along (“this is what you did last time, I need it to be better”). It never occurred to me that RC might be “lazy” and think “Hey, well if you liked this one last time,here it is again!”

Already had one HUGE error on a selected CP…but to be fair it was right at the edge of field of a GoPro…so, y;know…a bit on the “curvy” side…

BTW…there’s a setting that is “maximum reprojection error” What DOES RC do about control poitns that are WAY off…does it look at a 24 point error (against  a limit of 3.0) and think “Hey, I’m not touching that”…or rather…“Hmm, that’s BADLY wrong…I better think again”.

It would be nice if there was a User option to choose…I know we have the “weighting” thing, but there should be a way of saying…“Look RC, I know you can operate on trillions of points at a speed that I can’t match…but when I say this point HERE is the same as this point HERE…I bloody mean it!”.

I think we need that sort of “God Level” control points…but limited to a maximum of three per alignment…it would be like saying either…

“I don’t care about the rest, focus on this”

or

“I’m 100% certain that these points are the same point in real space”

Why three? welll…even given the last statment…I’m only human…and one block of stone can look very much like another when you zoom in and…I’m getting on a bit and my eyesight’s not great…and I’d had a few…and the cat was being sick…and…and…

Given all that, best not to present RC with too much “absolute certainty” that turns out to be “absolute garbage”…

…and…three points…recognised way of defining a 3D space - one in each plane…

Possible?

Tom, thanks for the flattery!  :slight_smile:

Nigel, it’s not really lazyness. In most cases, it is a huge advantage - mostly in terms of speed for later optimizations. The thing is that RC is a bit of a Diva - really good but sometimes tricky to handle. So you need how to learn to handle it (image-wise) and then the standard settings with slight tweaks are sufficient to get perfect results.

The problem with strongarming RC with CPs is that RC still needs tie points that it can find for itself. So if your opinion and RCs don’t go together, there is nothing you can do to change that. Fine-tuning yes, but not much more. So I really suggest to invest your time and effort in image taking rather than CP placement training.  :slight_smile:

“RC is a bit of a Diva - really good but sometimes tricky to handle”

More accuracy :slight_smile:

I’ll admit I’m currently at quantity over quality, and starting out trying to capture a 50m or so tower in stone by stone detail is perhaps a little…ambitious.

But I _will_ succeed…

to which end, when you say taking images…what do you mean…what is the best way to imrpove RC’s results…what does it look for in a “good” image?

For instance, I did wonder if “blur” would confuse it…am I better going through manually and just deleting any that have mation blur (from drone) or other out of focus (from expensive camera operated by blind old idiot) issues…?

I set out to answer but moved it to a new thread: https://support.capturingreality.com/hc/en-us/community/posts/360009255471-A-good-photo-set